Machine Learning Prediction Models For Classifying Myopic Versus Hyperopic Ablation Patterns In Eyes Undergoing Post-Refractive Cataract Surgery

To develop and validate machine learning models to classify prior myopic versus hyperopic ablation patterns in post-refractive eyes using corneal topographic and tomographic parameters.

Instituto de Oftalmología Conde de Valenciana, Mexico City, Mexico.

Seventy-two eyes with previous LASIK, PRK, or KLEx were analyzed. Predictive variables included corneal asphericity (Q value), posterior-to-anterior curvature ratio (P/A ratio), sagittal map morphology, spherical aberration (Z4⁰), mean keratometry, central corneal thickness, and thinnest pachymetry. Data were randomly divided into 70% training and 30% testing sets using Orange Data Mining software with stratified sampling. Random Forest, Decision Tree, and Logistic Regression models were trained and evaluated. Performance metrics included area under the ROC curve (AUC), accuracy, F1-score, and Matthews correlation coefficient (MCC).

Of 72 eyes, 60 (83.3%) had prior myopic and 12 (16.7%) hyperopic ablation. P/A ratio and Z4⁰ showed strong negative correlation (r = −0.793, p < 0.001). Random Forest demonstrated highest discriminative performance (AUC = 0.933; accuracy = 0.90), followed by Logistic Regression (AUC = 0.881; accuracy = 0.886) and Decision Tree (AUC = 0.828; accuracy = 0.90). Decision Tree identified clinically interpretable cutoffs: P/A ratio ≤ 82.7% indicating myopic ablation and Z4⁰ ≤ −0.144 indicating hyperopic ablation. Random Forest yielded the lowest misclassification rate (6.9%).

Machine learning models reliably distinguished prior myopic from hyperopic ablation patterns using corneal imaging parameters. Random Forest demonstrated superior performance, while Decision Tree provided clinically interpretable thresholds. These tools may support intraocular lens power selection in post-refractive cataract surgery when historical surgical data are unavailable.